Opting Out of AI Training in 2027 Without Disappearing From AI Answers

Two announcements on October 30, 2026 change how a photography site tells AI companies what they may do with it. Google, Microsoft, OpenAI and Anthropic backed a single shared line for robots.txt that says “don't train AI models on my site”. Separately, Bing is retiring the old tags some sites used to keep their pages out of Copilot's answers.

Both are good news if you've wanted a simple way to keep your photos out of training data. But the same line means different things to different companies. Add it without thinking it through and you can quietly drop out of Gemini's recommendations, which is the exact opposite of what most photographers want.

Training and answers are two different things

When an AI company uses your site, it's doing one of two jobs:

• Training: your pages and photos become part of the material a future model learns from. You get nothing back, and no one is sent to your site.
• Answers (the industry calls this grounding): someone asks “who's a good family photographer in Denver?”, the assistant reads your site right then, and names or links you in its reply. That's a referral.

Most photographers want to say no to the first and yes to the second. Whether you can depends on which company you're talking to.

The new shared opt-out

The new line is a normal robots.txt group with an unusual name:

User-agent: __usage_ai-training
Disallow: /

It's based on a proposal at the IETF, the body that maintains web standards. A rule for User-agent: * does not count, so an existing “block everything” rule won't opt you out by accident. Microsoft says it will start honouring the line on February 1, 2027, and Google is aiming for early 2027.

Where it goes wrong: Google

Microsoft reads the line as training-only. Your pages can still appear in Copilot answers.

Google is different. It has never had a switch that blocks training but allows answers. Both are controlled by one token, Google-Extended. Google has said that if your robots.txt has the new training line and no rule of its own for Google-Extended, it will treat that as blocking Google-Extended. That means no Gemini training, and also no Gemini answers that cite or recommend you.

If you want Gemini to keep recommending you, add a separate group that speaks to Google directly:

User-agent: __usage_ai-training
Disallow: /

User-agent: Google-Extended
Allow: /

The honest trade-off: this lets Google use your site for Gemini training too, because Google offers no way to separate the two. You're choosing between “Google may train on my work and Gemini can recommend me” and “neither”. There's no wrong answer, but it should be a choice, not an accident.

None of this touches ordinary Google Search, or AI Overviews and AI Mode. Those have their own setting in Search Console. See Google's AI opt-out setting.

Bing: NoArchive and NoCache stop working for AI on March 1, 2027

Some sites carry a noarchive or nocache robots tag, often added by a security or “protect my images” plugin rather than chosen. Bing has read these as “don't use this page in Copilot answers”. It's replacing them with separate AI training and AI grounding preferences, and after March 1, 2027 the old tags no longer count.

• If you want Copilot to be able to recommend you: do nothing. Pages that relied on the old tags default to allowed.
• If you really meant to keep Copilot out: set the new grounding preference on those pages. Also record a temporary preference in Bing Webmaster Tools (available from December 15, 2026) before March 1. It covers you until Bing recrawls the page.

Can you even edit robots.txt?

On self-hosted WordPress, yes: through an SEO plugin or your host. Hosted builders are more limited. Squarespace offers a toggle for AI crawlers rather than a free-form file, and some portfolio platforms give you no access at all. If you can't add the new line yourself, it's worth asking your platform's support whether they plan to support it. Platforms usually add these as a checkbox once the big companies start honouring them.

Whatever you decide, keep in mind what robots.txt is: a request that well-behaved crawlers choose to honour, not a lock. For what happens when a CDN does the blocking instead, see Cloudflare's AI bot change.

How to check yours

Every PhotoSEO audit reads your robots.txt and your pages' robots tags. It flags both situations above: a training opt-out that will also hide you from Gemini, and pages relying on NoArchive or NoCache for Bing. It also tests whether ChatGPT, Claude, Perplexity and Gemini can reach your site at all, and that is usually the bigger problem.